it does seem to be quite strongly encapsulated. Thus when watching a good
actor on the stage, for example, I cannot help but see his actions as
deceitful, jealous, angry, or whatever – despite knowing full well that he is
really none of these things.
To take another example, Cosmides and Tooby (1992) have argued
persuasively for the existence of a special-purpose ‘cheater-detection sys-
tem’. Since various forms of co-operation and social exchange have prob-
ably played a signiWcant part in hominid lifestyles for many hundreds of
thousands of years, it makes sense that these should have been underpin-
ned by a cognitive adaptation. This would operate upon conceptual in-
puts, analysing the situation abstractly in terms of a cost-beneWt structure,
so as to keep track of who owes what to whom, and to detect those who try
to reap the beneWts of co-operative activity without paying the costs.
Cosmides and Tooby also claim toWnd direct experimental evidence in
support of their proposal, deriving from subjects’ diVerential performance
in a variety of reasoning tasks (more on this in chapter 5).
It is important to see that thedomain-speciWcityof central modules is
consistent with their having a variety of diVerent input sources – perhaps
receiving, as input, suitably conceptualised outputs from many of the
various input-modules. For central modules may only operate upon inputs
conceptualised in a manner appropriate to their respective domains, such
as action-descriptions, in the case of mind-reading, or cost-beneWt struc-
tures, in the case of the cheater-detection system.
If there are similarly conceptualised forms of representation produced
as outputs by a variety of input-modules, then a module which took such
outputs as itsinput might appear to be less than fully domain-speciWc. But
such a modular central-system might in fact be thought of as having both a
fairly speciWc proximal domain (given by the terms in which its inputs are
conceptualised) and also a more general remote domain. Consider, for
example, the suggestion of a logic-module which processes simple inferen-
ces and checks for their validity (Sperber, 1997). These inferences could be
aboutanything at all. So such a system has a remote domain of con-
siderable generality. On the other hand, all it is doing is checking for
formal validity and consistency in the representations which it takes as
input. So its proximal domain would be quite limited and speciWc, being
conWned to a syntactic and lexical description of the input-representations.
Once we accept that a central module can take as input the output from
various peripheral modules, then it should not be diYcult to believe that it
might also be able to make use of things known to some other parts of the
cognitive system. In particular, it might be able to take as input the outputs
of various othercentralmodules (as when the output of the cheater-
detection system is fed into the mind-reading system, say, to work outwhy
68 Modularity and nativism